Tegra k1

[monstercameron]

http://www.gputechconf.com/page/home.html

so what do you think we will see?

 

[Arkadrel]

More of same.

A tegra k1 running 10-15 watts, beating the snot out of competitors (2w chips), while JHH neglects to mention his chip is running at that high power.

Then when it launches in smaller devices everyone will be like....

"what happend to all that performance we where shown in demos by jhh?"

 

[monstercameron]

More of same.

A tegra k1 running 10-15 watts, beating the snot out of competitors (2w chips), while JHH neglects to mention his chip is running at that high power.

Then when it launches in smaller devices everyone will be like....

"what happend to all that performance we where shown in demos by jhh?"

Well I am optimistic, This will be their early APU approach -Until we see that TK1V2. The gpu will be better than kabini with better power draw and maybe more cpu performance -obviously beema/mullins will correct that.

shield 2 seems very promising. we just need more game devs porting to opengl. On the topic of shield, paradox just ported mount & blade to android [tegrazoned], which is just pure epic.
https://www.youtube.com/watch?v=Fw7a6uMZaHI

 

[NTMBK]

Well I am optimistic, This will be their early APU approach -Until we see that TK1V2. The gpu will be better than kabini with better power draw and maybe more cpu performance -obviously beema/mullins will correct that.

shield 2 seems very promising. we just need more game devs porting to opengl. On the topic of shield, paradox just ported mount & blade to android [tegrazoned], which is just pure epic.
https://www.youtube.com/watch?v=Fw7a6uMZaHI

Is Shield 2 confirmed? The first one was a complete sales bomb...

 

[bullzz]

anything on tegra k1? thought it was more abt future products

 

[sontin]

More of same.

A tegra k1 running 10-15 watts, beating the snot out of competitors (2w chips), while JHH neglects to mention his chip is running at that high power.

Then when it launches in smaller devices everyone will be like....

"what happend to all that performance we where shown in demos by jhh?"

2W? You mean like all these chips which are throlling at 2W? :awe:

 

[Gloomy]

Is Shield 2 confirmed? The first one was a complete sales bomb...

lol, they're giving them away

 

[bullzz]

giving them away for attendees

but looks like there will be price cut
http://www.engadget.com/2014/03/25/nvidia-shield-gets-temporary-price-drop-soon-to-support-remote/

 

[erunion]

Semiaccurate noticed some roadmap changes

http://semiaccurate.com/2014/03/25/nvidia-just-cancel-project-denver-gtc/

 

[bullzz]

anand noticed the change in his live blog. was parker project denver?

 

[jdubs03]

On the slide presented showing the roadmap of Tegra and now Erista, it indicates Tegra K1 as 32b and 64b. The 64b is Denver. I don't think anything changed in that regard. Parker was listed as Maxwell GPU and finfet, the only difference here is the Erista doesn't have finfet listed, so that may be a 20nm process as opposed to TSMC's 16FF (16nm is likely to be in production late in 2015, so that explains that).

With pascal, there was a change from Volta, along with a reduction in gflops/w (not massive, but nonetheless a reduction).

edit: according to fudzilla, pascal got pulled in and volta is the successor.

http://www.fudzilla.com/home/item/34311-volta-coming-after-pascal-2016

 

[erunion]

ARM has 64 bit cores. 64bit doesn't prove it is Denver.

 

[sontin]

On the slide presented showing the roadmap of Tegra and now Erista, it indicates Tegra K1 as 32b and 64b. The 64b is Denver. I don't think anything changed in that regard. Parker was listed as Maxwell GPU and finfet, the only difference here is the Erista doesn't have finfet listed, so that may be a 20nm process as opposed to TSMC's 16FF (16nm is likely to be in production late in 2015, so that explains that).
http://www.fudzilla.com/home/item/34311-volta-coming-after-pascal-2016

Parker was announced with FinFet. I guess they are really pulling in a 20nm SoC and will bring Parker 2016.

 

[jdubs03]

ARM has 64 bit cores. 64bit doesn't prove it is Denver.

nvidia already showed the denver (64-bit) version of k1 at ces. as i said underneath the listing of tegra k1, it has 32b and 64b, we were referring to that on these forums as k1v1 and k1v2, that is cortex a15 r4 for v1 and denver for v2. there should be no confusion on this subject. i doubt that they would scrap a 5+ year development that quickly.

it is pretty safe to assume that any future armv8 core will be an evolution of the denver core.

Parker was announced with FinFet. I guess they are really pulling in a 20nm SoC and will bring Parker 2016.

that would make sense, and backs up my point, with tsmc going into 16ff production in late 2015 it would be impossible for nvidia to release parker with finfets until 2016. (my guess is they may rename parker to something else to not make it seem like a roadmap slip)

 

[monstercameron]

nvidia already showed the denver (64-bit) version of k1 at ces. as i said underneath the listing of tegra k1, it has 32b and 64b, we were referring to that on these forums as k1v1 and k1v2, that is cortex a15 r4 for v1 and denver for v2. there should be no confusion on this subject. i doubt that they would scrap a 5+ year development that quickly.

it is pretty safe to assume that any future armv8 core will be an evolution of the denver core.



that would make sense, and backs up my point, with tsmc going into 16ff production in late 2015 it would be impossible for nvidia to release parker with finfets until 2016. (my guess is they may rename parker to something else to not make it seem like a roadmap slip)

can you provide a source where nvidia confirmed denver in project logan? I believe 64 bit refers to non-denver armv8 cores.

 

[erunion]

nvidia already showed the denver (64-bit) version of k1 at ces. as i said underneath the listing of tegra k1, it has 32b and 64b, we were referring to that on these forums as k1v1 and k1v2, that is cortex a15 r4 for v1 and denver for v2. there should be no confusion on this subject. i doubt that they would scrap a 5+ year development that quickly.

The old roadmap specifically said maxwell and denver in 2015. The new roadmap only says maxwell.

Obviously, something has changed since CES.

 

[erunion]

Parker was announced with FinFet. I guess they are really pulling in a 20nm SoC and will bring Parker 2016.

That seems to be the cause of the roadmap changes. Products using TSMC 16nm are no longer on it. The new 2015/2016 products are likely on some form of TSMC 20nm.

That begs the question, whats happening with TSMC 16nm?

 

[Exophase]

The old roadmap specifically said maxwell and denver in 2015. The new roadmap only says maxwell.

Obviously, something has changed since CES.

They're introducing Denver on the second iteration of K1, allegedly a 2014 part - we'll see if they keep to that schedule. They list underneath the roadmap item newly introduced technologies. This is what they've always done. Denver wouldn't be new in Erista (although that doesn't mean they'll definitely use it, it at least doesn't suggest they won't)

Nothing's changed since CES. They've been talking about Denver for years, filed patents regarding the technology, released further information and announced an SoC that'll use it only months away. The idea that they threw this all out and replaced it with another 64-bit ARM core a mere months before release does not hold water.

And I don't know why people are making such a huge fuss over nVidia's roadmap changing when it was projected several years ahead. It was already evident they changed it with K1v2 which is clearly not Parker (no Maxwell GPU, no FinFETs) Just another case of Charlie terribly misinterpreting things.

 

[erunion]

They're introducing Denver on the second iteration of K1, allegedly a 2014 part - we'll see if they keep to that schedule. They list underneath the roadmap item newly introduced technologies. This is what they've always done. Denver wouldn't be new in Erista (although that doesn't mean they'll definitely use it, it at least doesn't suggest they won't)

That's a reasonable explanation.

 

[witeken]

can you provide a source where nvidia confirmed denver in project logan? I believe 64 bit refers to non-denver armv8 cores.

They announced Tegra K1 (aka Tegra 5 aka Logan) at CES. They announced 2 versions: a 32-bit version with regular A15 cores and a 64-bit Denver dualcore.

Tegra K1 with Denver is something that Nvidia didn't announce before CES:

 

[cbn]

This just showed up in Newegg embedded solutions today:

http://www.newegg.com/Product/Produc...82E16813190005

(Release data 4/24/2104 and $192.99 Pre-order)

Dimensions 5" x 5"

Nvidia K1 processor (4+1 Cortex A15 plus 192 CUDA core Kepler GPU)
2GB RAM
16GB eMMC 4.51
1 x Half mini-PCIE
1 x SATA
Realtek ALC5639 ausi chipset
Realtek RTL8111GS LAN chipset
HDMI
1 x usb 3.0
2 x audio ports
RS232 (serial port)
1 x Full size SD/MMC connector
1 x JTAG
DC-in power

Furthermore, it mentions the The following signals are available through an expansion port:
-DP/LVDS
-Touch SPI 1x4 + 1x1 CSI-2
-GPIOs
-UART
-HSIC
-i2c

Also mentioned is a "Linux for Tegra" Operating system clearly aimed at Developers. (see the Newegg link for more info)

 

[monstercameron]

This just showed up in Newegg embedded solutions today:

http://www.newegg.com/Product/Produc...82E16813190005

(Release data 4/24/2104 and $192.99 Pre-order)

Dimensions 5" x 5"

Nvidia K1 processor (4+1 Cortex A15 plus 192 CUDA core Kepler GPU)
2GB RAM
16GB eMMC 4.51
1 x Half mini-PCIE
1 x SATA
Realtek ALC5639 ausi chipset
Realtek RTL8111GS LAN chipset
HDMI
1 x usb 3.0
2 x audio ports
RS232 (serial port)
1 x Full size SD/MMC connector
1 x JTAG
DC-in power

Furthermore, it mentions the The following signals are available through an expansion port:
-DP/LVDS
-Touch SPI 1x4 + 1x1 CSI-2
-GPIOs
-UART
-HSIC
-i2c

Also mentioned is a "Linux for Tegra" Operating system clearly aimed at Developers. (see the Newegg link for more info)

I am in support of nvidia's tegra but why would anyone buy this compared to a baytrail board or a am1 board? atleast for the price [hey look at that $192, I wonder what is the significance? lol], do embeded developers really want cuda?

 

[NTMBK]

I am in support of nvidia's tegra but why would anyone buy this compared to a baytrail board or a am1 board? atleast for the price [hey look at that $192, I wonder what is the significance? lol], do embeded developers really want cuda?

Useful to verify that you can get your CUDA code running on an ARM Linux distro; a good first step, if you want to run it on any potential future NVidia ARM CPUs. (Although the Denver version will be more useful.)

 

[cbn]

I am in support of nvidia's tegra but why would anyone buy this compared to a baytrail board or a am1 board? atleast for the price [hey look at that $192, I wonder what is the significance? lol]

I do agree that it is totally overpriced for consumer.

With that mentioned, I do like some of the features it has such as soldered on 2GB RAM, 16GB eMMC (in addition to SATA port) and DC-in and wish those would also make it onto future extreme budget x86 Mini-ITX boards (particularly when eMMC 5.0 comes out).

 

[Vesku]

It's a development board, basically, wish more of this type of stuff would be sold on the big e-tailer sites. $193 is much better than the $300-500 you'd expect to find something like this on specialist sites that stock dev kits.